Multiscale network alignment model based on convolution of homogeneous multilayer graphs

Social network alignment as an important research method in network science has been widely used in several fields. Existing methods usually rely on high-quality user attribute information to complete specific tasks, but the existence of privacy protection mechanisms makes this information difficult to obtain. In addition, relying solely on network topologies can be challenged by insufficient data. In order to solve the above problems, a cross-network user alignment model based on the node neighborhood characteristics and network homogeneity was proposed. In terms of node characteristics, the K-nearest neighbor algorithm was used to aggregate node neighborhood information to model the deep network structure, so as to enhance the data. In terms of graph convolution, the convolution process was guided by the construction of a homogeneity matrix according to the network homogeneity, and the social networks of different scales were processed based on the network community structure. Experimental results on two real-world social networks of different scales show that the proposed method can effectively improve the performance of social network alignment tasks.